The heart of tech is coming to the heart of the Mediterranean. Join TNW in València this March 🇪🇸

This article was published on March 10, 2016

Shutterstock now uses machine learning to help you find just the right photo

Napier Lopez
Story by

Napier Lopez


Napier Lopez is a writer based in New York City. He's interested in all things tech, science, and photography related, and likes to yo-yo in Napier Lopez is a writer based in New York City. He's interested in all things tech, science, and photography related, and likes to yo-yo in his free time. Follow him on Twitter.

Shutterstock is today launching a couple of tools that will make it easy for users to find photos without having to figure out the right keywords for what they’re looking for: reverse image search and visually similar search.

It’s pretty much exactly what it sounds like. With reverse image search, you can upload an image that you like via the search bar and the site will pull up a collection of images that have a similar aesthetic.

Visually similar search uses the same technology, except it’s meant to provide you with more options as you are browsing Shutterstock’s own catalog. And while Shutterstock already provided a ‘Similar Images’ section, those were based off of keywords that often did not properly describe the image. For example, here are the results from old keyword based search:


And here are the results using visually simlar search.


Both features make use of a custom-built deep learning convolutional neural network, which is a fancy way of saying it uses advanced AI, similar to the technology behind Google’s DeepMind.

That said, Shutterstock wants to emphasize that its network is custom-built; while other stock image companies have release reverse image search tools before, Shutterstock says it waited to make sure it got its own right.

The company demoed the tool for me in their New York offices, including some of the multiple iterations the software had to go through before it began providing results with an accuracy the company was happy with.

For instance, earlier iterations would often pull up images with the same colors or patterns, but completely different subjects. Others sometimes got the subject right, but didn’t quite nail the ‘feel’ of the image, which is important for Shutterstock users looking to use photos for their own publications.

These were similar to the results from competing services, which often got the colors of the image right, but completely missed on the subject matter. Compare that to the current tool, and the results are night and day.

Finding an image that looks exactly like this cathedral shot through just keywords would be very difficult

It’s altogether a step forward from old keyword-based image search, where the combination of improperly tagged images and our inability to articulate exactly what we’re looking for can lead to many useless results.

But perhaps as importantly, Shutterstock says it’s already considering new ways to use the technology it’s built -video search is already on the way, and perhaps one day Shutterstock will be able to automatically tag your photos with proper keywords. In any case, reverse image search is only just the start.

Shutterstock Unveils Better, Faster, Stronger Search and Discovery Technology [Shutterstock Blog]